On 2015 Sep 11, David Reardon commented:

1. With only 44% of the expected rate for abortions in this sample, the concealment rate is very high and means that many women with a history of abortion are misclassified by the authors into the control group (delivering women with "no history of abortion"). Given the diluting effect this would have on results, this suggests that negative findings (lack of statistically significant differences) really don't tell us much. On the other hand, positive findings (statistically significant results) are most likely truly significant since the differences must be pronounced enough to still show up despite the dilution effect caused by concealment and misclassification.

2. Even in Model 4, with all the controls in place, higher rates of subsequent mental illness (RR>1) are seen for women admitting a history of abortion in every classification examined. Given the relatively small sample size, the lack of statistical significance in these cases can very likely be due to small sample size (and the concealment/misclassification problem) rather than lack of any true association.

3. It is unfortunate, and inappropriate, that the authors used history of miscarriage as a control variable on several counts. At the very least, women responding to surveys of this type frequently will disguise a past abortion by describing it as a miscarriage. In addition, abortion may contribute to a miscarriage. At the very least, the authors should reanalyze the data to show us how the results would differ if women reporting a first miscarriage were removed from the sample, which is the only way to properly eliminate any confounding effects that may be associated with miscarriage.

4. The authors chose to control for the number of pre-abortion psychiatric episodes/diagnoses, but they failed to explore whether abortion is associated with an increased rate in the number of episodes subsequent to the event. In other words, the authors made the odd decision to consider the frequency of prior psychiatric episodes but then failed to explore the concern that the intensity and frequency of subsequent mental health problems. If women with prior mental health problems are more likely to have abortions, the question then turns to whether or not having an abortion worsens, decreases, or has no effect on their mental health.

Given the weakness of this study, the authors' conclusions that this study suffices to demonstrate that other studies (including large record based studies) showing associations between abortion and mental illness can be ignored is imprudent and very likely engaging in a Type II error. Record linkage studies which have no dropout problem, have shown that the elevated rates of both inpatient and outpatient psychological treatments following abortion cannot be explained by recent mental health history alone. Similarly, an examination of attempted suicide rates before a pregnancy and after pregnancies ending in abortion, miscarriage, or delivery, that there is a true elevated rate of suicide attempts following abortion while the risk declines after miscarriage or delivery.

Moreover, given the weaknesses inherent in this data set, it is especially concerning that the authors dismissed their own finding that substance use problems remained significantly associated with abortion despite all the controls used which were specifically chosen to reduce significance. Instead of recommending caution and more research, they show their bias by arguing that this outcome, too, could probably be reduced to statistical insignificance if only they were able to find yet more controls to apply to the problem. This assertion, again, ignores the self reports of women who say their abortion loss contributed to their drinking or substance use, or studies showing elevated rates of substance us among women with no prior history of substance use, and despite major reviews of the literature and a meta-analysis showing a consistent link between abortion and elevated rates of substance use.Coleman PK, 2011

In short, the authors rush to dismiss a large body of evidence, including the self reports of self-aware women, suggesting that abortion contributes to mental health problems, based on this analysis of a flawed data set and questionable methodological choices, reaches way too far.

A national longitudinal study designed to specifically explore women's mental health in relation to their reproductive health and experiences is long overdue. The National Comorbidity Survey-Replication data set was not designed for this task and it is a mistake to conclude too much from any analysis of this type.

On 2015 Sep 11, David Reardon commented:

1. With only 44% of the expected rate for abortions in this sample, the concealment rate is very high and means that many women with a history of abortion are misclassified by the authors into the control group (delivering women with "no history of abortion"). Given the diluting effect this would have on results, this suggests that negative findings (lack of statistically significant differences) really don't tell us much. On the other hand, positive findings (statistically significant results) are most likely truly significant since the differences must be pronounced enough to still show up despite the dilution effect caused by concealment and misclassification.

2. Even in Model 4, with all the controls in place, higher rates of subsequent mental illness (RR>1) are seen for women admitting a history of abortion in every classification examined. Given the relatively small sample size, the lack of statistical significance in these cases can very likely be due to small sample size (and the concealment/misclassification problem) rather than lack of any true association.

3. It is unfortunate, and inappropriate, that the authors used history of miscarriage as a control variable on several counts. At the very least, women responding to surveys of this type frequently will disguise a past abortion by describing it as a miscarriage. In addition, abortion may contribute to a miscarriage. At the very least, the authors should reanalyze the data to show us how the results would differ if women reporting a first miscarriage were removed from the sample, which is the only way to properly eliminate any confounding effects that may be associated with miscarriage.

4. The authors chose to control for the number of pre-abortion psychiatric episodes/diagnoses, but they failed to explore whether abortion is associated with an increased rate in the number of episodes subsequent to the event. In other words, the authors made the odd decision to consider the frequency of prior psychiatric episodes but then failed to explore the concern that the intensity and frequency of subsequent mental health problems. If women with prior mental health problems are more likely to have abortions, the question then turns to whether or not having an abortion worsens, decreases, or has no effect on their mental health.

Given the weakness of this study, the authors' conclusions that this study suffices to demonstrate that other studies (including large record based studies) showing associations between abortion and mental illness can be ignored is imprudent and very likely engaging in a Type II error. Record linkage studies which have no dropout problem, have shown that the elevated rates of both inpatient and outpatient psychological treatments following abortion cannot be explained by recent mental health history alone. Similarly, an examination of attempted suicide rates before a pregnancy and after pregnancies ending in abortion, miscarriage, or delivery, that there is a true elevated rate of suicide attempts following abortion while the risk declines after miscarriage or delivery.

Moreover, given the weaknesses inherent in this data set, it is especially concerning that the authors dismissed their own finding that substance use problems remained significantly associated with abortion despite all the controls used which were specifically chosen to reduce significance. Instead of recommending caution and more research, they show their bias by arguing that this outcome, too, could probably be reduced to statistical insignificance if only they were able to find yet more controls to apply to the problem. This assertion, again, ignores the self reports of women who say their abortion loss contributed to their drinking or substance use, or studies showing elevated rates of substance us among women with no prior history of substance use, and despite major reviews of the literature and a meta-analysis showing a consistent link between abortion and elevated rates of substance use.Coleman PK, 2011

In short, the authors rush to dismiss a large body of evidence, including the self reports of self-aware women, suggesting that abortion contributes to mental health problems, based on this analysis of a flawed data set and questionable methodological choices, reaches way too far.

A national longitudinal study designed to specifically explore women's mental health in relation to their reproductive health and experiences is long overdue. The National Comorbidity Survey-Replication data set was not designed for this task and it is a mistake to conclude too much from any analysis of this type.